Discrete Time Multistate Model With Regime Switching for Modeling COVID-19 Disease Progression and Clinical Outcomes

Abstract

Randomized and controlled clinical trials are an important tool to provide definitive evidence about effects of treatments for COVID-19. Because of the novelty of this disease, designing these trials proves to be challenging absent reliable data on the interplay between disease severity, clinical course of disease and their impact to endpoints. A clinically meaningful endpoint and its associated effect size are needed to ascertain adequate sample size for the trial to meet its objectives. To gain insight into these questions, one can only rely on properly calibrated disease progression model to generate realistic daily patient clinical status. This is achieved through a discrete time multistate model with regime switching where the states are defined by the levels on the ordinal scale. The switch ensures that a patient on the way to recovery is not reinfected which provides a break to the memory-less property of traditional multistate models. The data obtained from this model provides insight on the impact of disease severity on operating characteristics of endpoints derived from the ordinal scale. In particular, concerns about power loss when dichotomizing the ordinal scale may not be dramatic compared to the analysis of the complete ordinal scale via proportional odds model.